[Mlir-commits] [mlir] [mlir] Fix semantics of linalg::ReduceOp with several inputs (PR #107005)

[llvmlistbot at llvm.org]

MaheshRavishankar wrote:

> > We can have multiple inputs and multiple outputs for a reduction. Thats totally OK, but you are saying its one op running independent reductions to generate different outputs, but with the same iteration space and dependence? That works.
> 
> Could you clarify what you mean here? I'm not sure I follow. The main point of this change is to be able to reduce several tensors into one, as in my examples, so the reductions are not independent, rather it uses all inputs.

Sorry to reverse the question on you. Can you write the loop version or  `linalg.generic` version of what the `reduce` is doing? What are the constraints on the shapes of the inputs? Should they all be the same shape? Are they all accessed using the same indexing map and is this checked in the verifier? 

https://github.com/llvm/llvm-project/pull/107005